基於擴展卡門濾波同時定位與地圖建立之地圖接合研究

Abstract

本論文提出一使用Kinect深度攝影機之機器人定位方法。以Kinect為感測器取得環境資訊，結合Extended Kalman Filter(EKF) 之同時定位與環境地圖建立(Simultaneous Localization and Mapping, SLAM)演算法，並以地圖接合之方式降低定位系統運算複雜度。Kinect深度影像攝影機同時提供彩色影像與距離資訊，本研究以擷取SURF特徵點對應感測器所取得之深度關係，快速且精確的取得特徵點之環境資訊，接著以EKF修正機器人狀態與特徵點三維座標。為了避免EKF隨環境增長而使矩陣運算複雜度快速增加，本論文提出以區域路徑範圍判斷之作法將環境分為數個子區域，機器人僅需要使用區域內之特徵點訊息，而無須使用整個環境資訊做定位系統演算，如此提高定位系統於機器人應用之即時運算之性能，最後鄰近的子區域以地圖融合演算法修正其間之差異，以維持整個環境地圖之完整性。實驗結果顯示機器人運行於一16mX7m之室內環境行走約83公尺，當機器人回到原點附近時實際位置與估測之間的二維座標平均誤差小於0.1公尺。實驗結果證實機器人能以EKF之定位系統藉地圖接合之作法，達成機器人室內導航之功能。

This study investigates simultaneous localization and mapping(SLAM) of a mobile robot using a Kinect depth camera. Depth and image information from Kinect are utilized to realize SLAM algorithms based on extended Kalman filter(EKF). In this thesis, visual landmarks are extracted by SURF algorithm, then three dimensional location of feature points are calculated from Kinect depth image data. A map joining method is proposed to reduce computational complexity of EKF-SLAM, and to correct the deviations of adjacent local maps. A global map of the environment is constructed by the map joining procedure. Navigation experiments show that the accuracy of robot localization for a travel about 83m path is within 0.1m. It is verified that the developed algorithm of simultaneous localization and mapping with map joining can allow robot to navigate in an indoor environment